Design and Fabrication of Crossing-Free Waveguide Routing Networks Using a Multi-Layer Polymer-Based Photonic Integration Platform

A novel 16×4 crossing-free waveguide routing network on four layers of polymer-based stacked waveguides is presented. The design combines in-plane passive waveguide devices with vertical multimode interference couplers (1×1 3D MMIs) to connect two adjacent waveguide layers. The 16×4 multilayer waveguide routing network (MWRN) is fabricated using multilayer deposition and standard UV contact lithography. A novel fabrication method of multiple waveguide layers that includes a sequence for wafer planarization steps to reduce topographies is presented. These developments lead to a reproducible multi-layer device fabricated on wafer-scale. Our technology enables a high vertical spacing of 21.6 μm (center-to-center) between the top and bottom waveguide layers with a large waveguide-to-waveguide spacing of 7.2 μm (center-to-center). The complex 16×4 MWRN can serve as the crossing-free waveguide routing for a 4×4 3D optical switch.

waveguide networks usually results in a significant number of waveguide crossings.These scales faster than proportionally with the number of ports.These crossings create crosstalk between the waveguides, which should be kept to a minimum.To reduce the number of waveguide crossings and the associated crosstalk, a new degree of freedom in the design of photonic devices is needed.This can be enabled by 3D photonic integration in the vertical direction.
Compared to the aforementioned platforms, the polymerbased photonic integrated platform PolyBoard presented here is also fabricated using multilayer deposition and UV lithography [25], [26], [27], [28], [29].Based on previous work on stacked and vertically connected waveguide layers [27], a new multilayer waveguide routing network (MWRN) was developed.This MWRN connects four vertical stacked waveguide layers and has a crossing-free waveguide design (Fig. 1).To the best of our knowledge, this device is the first to be fabricated with multilayer deposition and UV lithography with four vertically optically connected waveguide layers and has a combination of vertical coupling elements and in-plane waveguide structures on the different waveguide layers.The 16×4 MWRN (Fig. 1) enables a crossing-free waveguide routing that can be used as part of a 4×4 optical switch (Fig. 1).The contribution of the 3D PolyBoard to this optical switch matrix is to provide a crossing-free routing component (Fig. 1).An optical 4×4 switch would require an additional PIC with the corresponding switching components, as shown in Fig. 1.By separating the two parts of the PIC, the advantages of e.g., two different platforms can be beneficially combined.Both devices are designed in such a way that they are hybrid integrated into a crossing-free 4×4 optical switch matrix.
For MWRN, the vertical coupler type is a crucial building block.To this end, adiabatic tapers [7], [8], grating couplers [30], [31], or directional couplers [10] are often used in multilayer PICs.In the aforementioned types of coupling, evanescent coupling between waveguide layers is an important issue because the vertical distances between the waveguide layers are smaller (0.2 µm up to 3.7 µm) [7], [8], [10], [30], [31].To reduce unwanted evanescent coupling between the waveguide layers, a significant vertical distance between the layers is desirable.This requirement is met by polymer-based 3D MMIs with a vertical gap between two waveguide layers of 7.2 µm (center-to-center) [27], as shown in Fig. 1.The vertical low-loss connection between two waveguide layers with an 1×1 3D MMI has already been successfully demonstrated in the PolyBoard platform [27].Furthermore, these structures are easily cascadable [29], which is necessary for the intended complex routing design (Fig. 1).The design and functionality of the 16×4 crossing-free MWRN are the focus of this work.The necessary adaptation of the process to fabricate such a 3D device is clarified and the results emerging from the characterization are discussed.

II. DESIGN AND FUNCTIONALITY OF THE CROSSING-FREE WAVEGUIDE ROUTING
The target is to develop a crossing-free MWRN (Fig. 1) to connect 16 input waveguides to 4 output waveguides.The 16 input waveguides are given by the interface to the switching component (Fig. 1).A bundle of four waveguides belongs to one input waveguide of the 4×4 optical switch (Fig. 1).The waveguide bundle of four waveguides must now be equally distributed to the four outputs of the MWRN without physically crossing other waveguides.For this merging, planar passive structures, as shown in Fig. 1, are necessary in addition to the vertical coupling elements.Already included in the standard portfolio of the PolyBoard platform are the single-mode polymer waveguides with a cross section of 3.2 µm by 3.2 µm at a refractive index contrast of Δn = 0.03 and a wavelength of λ = 1550 nm with horizontal tapers and the S-shaped waveguide bends with a bending radius of 1500 µm.In addition, planar 4×1 MMIs are commonly used on the platform.The in-plane structures appear in the novel design in Fig. 1 on all four waveguide layers.The 4×1 MMI is used to combine the four waveguides per layer into one output waveguide and has a width of 54 µm, a length of 729 µm, and has been optimized to a wavelength of λ = 1550 nm.
Besides the in-plane waveguide structures, the 3D basic building blocks are to be considered as well.The vertical connection between two waveguide layers is achieved by an 1×1 3D MMI (Fig. 1).The single-mode input waveguides connect to the single-mode output waveguides in a higher waveguide layer via a vertical multi-mode section.Due to the multi-mode region between the waveguide layers (length of 173 µm, width of 1.4 µm), the light is coupled vertically.With a refractive index contrast Δn = 0.03 at a wavelength of 1550 nm, a vertical distance of 7.2 µm (center-to-center) can thus be bridged.In previous work, it was shown that vertical transmission was achieved with an excess loss of 1 dB, regardless of polarization [27].In another work, a vertical distance of 21.6 µm was bridged with an on-chip loss of 2.5 dB by cascading three 1×1 3D MMIs [29].The performance influences of the 1×1 3D MMI parameters length, width and height, as well as the consideration of the technology tolerance, are discussed in [27], [28], [29] in detail.
The basic building blocks listed up to this point (multi-layer waveguide, taper, bends, 4×1 MMI, and 1×1 3D MMI) form the basis for the 16×4 crossing-free 3D waveguide routing (Fig. 1).The schematic waveguide routing in Fig. 1 shows a staged routing from the lowest waveguide layer (blue) to the topmost fourth waveguide layer (red).This stepped design results in a small footprint and an even distribution of losses across all optical paths.
The optical path is briefly illustrated with an example.Every third waveguide from the four groups of input waveguides Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.(Fig. 1) is first routed from the first waveguide layer to the second waveguide layer through an 1×1 3D MMI (Fig. 1).There, the four paths are combined with a 4×1 MMI into one output waveguide (Fig. 1).Since all output waveguides should end at the same height, the waveguides are routed into the fourth layer after the 4×1 MMI with two additional cascaded 1×1 3D MMIs (Fig. 1).The output waveguides on the top waveguide layer at the chip edge are arranged to correspond to an optical waveguide array with a standard spacing of 250 µm (center-to-center).Hence, each optical path contains three 1×1 3D MMIs and one 4×1 MMI.

III. FABRICATION OF 3D POLYMER WAVEGUIDE STRUCTURES
Due to the layer structure, three layers must be fabricated for a single vertical connection, i.e., an 1×1 3D MMI as shown in Fig. 1 in the lower right box.The top and bottom layers are the waveguide layers that are vertically connected by the vertical MMI area.The middle layer is the vertical MMI layer.The layer system must be fabricated in such a way that the three layers are in direct contact with each other in the vertical MMI area.
This has already been successfully demonstrated for one 1×1 3D MMI as part of various polymer-based 3D devices [27], [28], [29].For a design with N vertically stacked waveguide layers to be vertically connected, 2N-1 core layers are required.Thus, the design from Fig. 1 consists of seven polymer layers.With this high number of layers, wafer topography has a strong influence on fabrication quality and reproducibility.Additional topo-graphy leads to layer thickness variation across the wafer.As a result, the waveguides themselves may vary in height or may not be processable at all.
Strong topography may arise during wafer processing, which is then propagated and reinforced in the further fabrication steps.Three main causes of topography are: i) thickness variations of the substrate, ii) inhomogeneity in the deposited polymer layer, and iii) the topography of the structured waveguides.
4-inch silicon wafers with a specified thickness of (525 ± 25) µm [32] are used as substrate.Due to the high manufacturing standard (SEMI standard) for such silicon substrates, these topographies are negligible compared to the other mentioned issues.The thickness of the polymer layers can also vary due to the deposition process (Fig. 2(i)).In this step, the liquid polymer material is applied to the wafer with a pipette and then distributed by a spin-coating process.Due to viscosity and surface tension, elevations are created mainly in the center and at the edge of the wafer.If the waveguides are structured, they lie open on the surface and form small topographies (Fig. 2(v)).By applying the next polymer layer, these waveguides are encapsulated, but locally elevations remain in the layer (Fig. 2(vi)).These local topographical structures lead to an impairment of processability due to spreading and amplification in the process.
To reduce and avoid the topographies, planarization steps are added to the standard multilayer manufacturing process.The step sequence of the process to fabricate a waveguide or MMI layer is shown in Fig. 2. Two optical polymers are used to make a single-mode waveguide.They differ only in refractive index (n core = 1.48; n clad = 1.45) (ChemOptics, [33]).First, the liquid cladding polymer is applied to the substrate by spinning, cured by UV irradiation, and provided with a backing.This cladding layer is mechanically planarized with a fly-cutter (Disco Corp., [34]).A diamond bit is used to remove the protruding elevations and create a defined surface level (Fig. 2(ii) and (vii)).This process creates a roughness on the surface of the cladding layer.To compensate for this roughness, a very thin additional layer of cladding polymer (< 2 µm) is applied.
Only the cladding is suitable for planarization, as significantly higher propagation loss would occur in the waveguide core due to the roughness and the additional interface.In addition, the significantly thicker cladding layers lead to higher topographies than in the core layer and are therefore of greater concern.
The core material is then applied to the thin cladding layer (Fig. 2(iv)).For the structuring of the core layer, the waveguide mask is transferred to the layer by patterning a titanium (Ti) hard mask using standard photolithography.The layer is structured by oxygen plasma etching, which transfers the mask to the layer (Fig. 2(v)).A unique feature of the fabrication of the 3D multilayered waveguides is that once the waveguide is patterned, the Ti mask remains in place for the next steps [28].Subsequently, another layer of cladding polymer is applied.The cladding completely encapsulates the waveguide (Fig. 2(vi)).Subsequently, this cladding is also planarized (Fig. 2(vii) and (viii)).To enable direct contact with the adjacent waveguide layer, the cladding is removed to expose the top edge of the lower waveguide by oxygen plasma etching.The Ti mask protects the waveguide during this process step.When the top edge of the waveguide is exposed again, the Ti mask is removed using hydrofluoric acid (Fig. 2(ix)).Despite the reduction of topographies, the vertical position of the waveguide top edge varies.To achieve complete exposure of the Ti mask, a slight over etch of 500 nm to 800 nm is required.Due to the oxygen plasma etching process, the side edges are slightly etched as well.In areas of the single-mode waveguide, this change in waveguide width is negligible as it does not affect the single-mode behavior.In the area of the 1×1 3D MMI, this over etch has no effect on performance, because core material is applied again in the next step and the MMI region is built up.Thus, the etched MMI region is enveloped by a core layer and the etched region is filled.
The sequence of steps is repeated for all waveguide and MMI layers.By reducing the topographies, there is no fundamental limit to this 3D integration technology with UV lithography in the vertical direction.The 16×4 crossing-free 3D waveguide routing chips could thus be reproducibly fabricated at waferscale.Fig. 3 shows a top view of such a chip.

IV. CHARACTERIZATION
In the last step, the fabricated waveguide routing network is optically characterized.For this purpose, the transmission is measured as a function of wavelength for each direct optical path of the 16 input and 4 output waveguides.For this optical characterization, an 1×1 3D MMI with the following device parameters was selected: MMI length L 3D MMI = 173 µm, MMI width W 3D MMI = 1.4 µm and vertical waveguide distance D 3D MMI = 7.2 µm (center-to-center), as this set of parameters shows the lowest losses for the 1×1 3D MMIs [27], [29] at 1550 nm the wavelength.Fig. 4(a) to (d) shows the insertion loss for the four outputs of the device as a function of wavelength for each of the four waveguide connections.All waveguide connections work.Noticeable are the higher losses for output O2 (Fig. 4(b)); these are the effects of a fabrication error where the compensation layer was omitted after planarization (Fig. 2(iii) and (vii)).The defect only occurs in the second waveguide layer.For the other waveguide outputs O1, O3, and O4, the insertion loss for the target wavelength 1550 nm is in the range of 10.2 dB to 11.3 dB (Fig. 4(a), (c), and (d)).
The present measurement results include the following losses: 4×1 inherent MMI loss of 6 dB, fiber-to-chip coupling loss, propagation loss, and design loss of the cascaded 1×1 3D MMIs.The loss values are summarized in Table I.The coupling and propagation losses were determined by measuring several reference waveguides on the three waveguide layers.Furthermore, the losses for the three times cascaded 1×1 3D MMIs result from the already known trade-off between a low-loss 1×1 3D MMI and a reproducible manufacturable design (Table I).This means that the structures are already known from the simulations of the cascaded 1×1 3D MMIs [29] to have these intrinsic losses Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.I).In the previous work [29], the influences of the parameter length and height on the cascaded 1×1 3D MMI structure could also be considered in more detail by measurement and simulation.From this, it could be shown that a deviation of the MMI length leads to higher losses.The influence of additional losses (Table I) originating from the in-plane 4×1 MMI is, nevertheless, is expected to be greater on the resulting additional losses per output.This is in contrast to the influence of the total MMI height, which leads to the shift of the central wavelength as observed in Fig. 4. From these measurements it can be inferred that only a slight deviation of the height below 400 nm [29] from the target value occurs.For the intrinsic losses given here, only the simulation result for cascaded 1×1 3D MMIs without deviations in the parameter were considered [29].After deducting the above-mentioned losses, the additional losses are in the order of 0.2 dB to 1.3 dB for outputs O1, O3, and O4.These may be due to additional propagation losses, as the optical path is longer than for the measured reference waveguides.
Furthermore, additional losses may be present due to the 4×1 MMIs and the 1×1 3D MMIs.For example, even small variations in the height of 1×1 3D MMI could shift the central wavelength from 1550 nm to other wavelengths [29].This leads to additional losses.Based on the measurement data in Fig. 4(a) to (d), it could be shown for the presented device that all waveguide connections between the sixteen input and the four output waveguides work.Only the values of the losses fluctuate for the reasons mentioned.Thus, to the best of our knowledge, a functioning 3D PIC with four waveguide layers in combination with in-plane passive structures has been fabricated using multilayer deposition and UV lithography.To assess the reproducibility of the fabrication process, several devices with the same set of parameters for the 1×1 3D MMIs were examined on two different wafers.For better comparability, both wafers examined have the discussed deviation in the fabrication steps in the second waveguide layer.There were three devices with the same parameter for the 1×1 3D MMIs per wafer.The three device positions, shown in Fig. 5, are constant between wafers.Fig. 5 shows the averaged insertion loss for the four outputs per device and wafer for a wavelength of 1550 nm.The higher losses at output O2 are obvious.Looking at the differences between the same devices and outputs in the average insertion loss, the maximum deviation is 1.0 dB.For all outputs, except output O2, and devices, the deviations are within a maximum of 1.5 dB.The components shown in Fig. 5 have been fabricated reproducibly.Thus, the adaptation of the fabrication process with regard to the planarization of the wafers has a positive effect on the yield of such 3D devices.
Thus, it could be shown that the novel multilayer structure of the crossing-free 16×4 waveguide routing is functional and reproducible and can be fabricated for different parameter sets for the 1×1 3D MMIs on wafer-scale with four waveguide layers.

V. CONCLUSION
For the first time, a functional 16×4 MWRN with four waveguide layers has been fabricated by multilayer deposition and standard contact lithography.The design of this device is based on waveguide routing for a crossing-free 4×4 optical waveguide switch and the separation of the switching area and waveguide routing.A crucial point is the combination of already-known in-plane structures with the vertical MMI coupling structures and the resulting crossing-free waveguide routing.
The design was successfully fabricated using an advanced fabricating process with planarization steps as a new advantageous sequence.Planarization aims to reduce the topographies already present on the substrate or created during the fabrication of the layered sequence.A fabricated device is shown in Fig. 3.The crossing-free 16×4 waveguide router was optically characterized.The transmission of all waveguide connections between the input and output waveguides was presented.From the results, it can be concluded that all connections and thus the entire 16×4 MWRN are functional.Considering the additional losses, it could be shown that these are in the order of max.1.3 dB for the presented device.Based on these measurement results, further 16×4 3D routings were characterized.The reproducibility of the fabrication process and results were investigated for six 16×4 MWRN devices with the same parameter set on two independent wafers.The crossing-free 16×4 waveguide router over four waveguide layers could be fabricated reproducibly on wafer-scale.This significant technology development leads to low additional losses and higher reproducibility.The four waveguide layers shown here are the starting point for further complex MWRN with waveguide layers ≥ 5 in the polymerbased integration platform.Furthermore, the novel combination of 1×1 3D MMIs and in-plane waveguide structures forms the basis for a new generation of passive devices without waveguide crossing and thus reduced crosstalk.
This work paves the way for new optical switching matrices with separate areas for switching elements and routing to take full advantage of different integration platforms.Additionally, the development of multilayer PICs in the PolyBoard platform opens further application areas, such as 3D optical phased arrays for wireless optical communication or low-loss multi-core fiber interconnects for PICs.

Fig. 1 .
Fig. 1.Scheme of a 4×4 3D optical switch (left) with separation of the active switching component on a planar PIC on the left and 16×4 crossing-free MWRN on a 3-dimensional polymer-based platform on the right.The 3D scheme on the right shows the 16×4 MWRN design in the detail.The waveguides of the same color are in the same waveguide layer.From the left side, light is coupled into sixteen waveguides (blue) through the routing structure to four output waveguides (red) on the right side.Below in the left box, an 1×1 3D MMI for the vertical coupling is shown as small rectangles and connection areas between the waveguide layers.The right box shows the longitudinal section of the 1×1 3D MMI with the representation of the different waveguide and vertical MMI layers.

Fig. 2 .
Fig. 2. Sequence of steps to fabricate a waveguide or MMI layer in polymer-based 3D integrated PICs.The standard process has been extended to include the steps of planarization and compensation for the resulting roughness.

Fig. 3 .
Fig. 3. Photo with inverted colors of the fabricated polymer-based 3D waveguide routing device (13 mm×5.5 mm) with the presented 16×4 MWRN design.On the left are the input waveguides I on the lowest waveguide layer.The output waveguides O on the right side are color coded (blue = lowest layer, ... red = highest layer), depending on which waveguide layer the planar 4×1 MMI is located on.

Fig. 4 .
Fig. 4. Insertion loss as a function of wavelength for each input-output waveguide combination of the 16×4 3D waveguide routing design.For each output waveguide (a)-(d), the four connected input waveguides (I) were measured.The insertion losses include 4×1 inherent MMI loss of 6 dB, coupling losses, and all on-chip losses.

Fig. 5 .
Fig. 5. Average insertion loss per waveguide output (O) for different identical 16×4 3D PICs.The results are broken down by output waveguide O1 (red) to O4 (blue).In addition, the PICs considered were on two different wafers (wafer 1 dots, wafer 2 crosses) to consider the reproducibility of the measurement results for devices with the same parameter set.

TABLE I COMPOSITION
OF THE LOSSES FOR 16×4 MWRN WITH FOUR WAVEGUIDE LAYERS AT THE WAVELENGTH OF 1550 NM (Table